A Novel Method to Improve the Visual Quality of X-Ray CR Images

Home , Zhengzhou University

I.J. Image, Graphics and Signal Processing, 2011, 4, 25-31 Published Online June 2011 in MECS (http://www.mecs-press.org/)

A Novel Method to Improve the Visual Quality of X-ray CR Images

Huiqin Jiang School of Information Engineering and Digital Medical Image Technique Research Center Zhengzhou University, Zhengzhou, China Email:[email protected]

Zhongyong Wang School of Information Engineering and Digital Medical Image Technique Research Center, Zhengzhou University, Zhengzhou, China Email:[email protected]

Ling Ma Fast Corporation 2791-5 Shimoturuma Yamoto, Kanagawa Japan. Email: [email protected]

Yumin Liu Business School and Digital Medical Image Technique Research Center , Zhengzhou University, Zhengzhou, China Email:[email protected]

Ping Li School of Information Engineering and Digital Medical Image Technique Research Center, Zhengzhou University, Zhengzhou, China Email: [email protected]

Abstract—The aim of this study is to improve the visual quality of x-ray CR images displayed at general displays. I. INTRODUCTION Firstly, we investigate a series of wavelet-based denoising methods for removing quantum noise remains in the As computer technologies advanced and imaging original images. The denoised image is obtained by using modalities allowed acquisition of digital images, Picture the scheme of wavelet thresholding, where the best suitable Archiving and Communications Systems (PACS) began wavelet and level are chosen based on theory analysis. to rapidly develop. Nowadays, many hospitals are Secondly, the image contrast is enhanced using Gamma undering the clinical operation using filmless full PACS correction. Thirdly, we improve unsharp masking method and many diagnoses are being made based on medical for enhancing some useful information and restraining other information selectively. Fourthly, we fuse the denoised softcopy image display system rather than conventional image with the enhanced image. Fively, the used display is film-on-light box display. However, in comparison with calibrated, so that it could offer full compliance with the the latter, medical softcopy image display may Grayscale Standard Display Function (GSDF) defined in compromise image quality [1], and some unexpected Digital Imaging and Communications in Medicine (DICOM) effects on the accuracy of diagnosis may be introduced. Part 14. Finally, we decide parameters of the image fusion, Also, there are many other issues like the viewer soft resulting in the diagnosis image. A number of experiments related to display quality unsettled in medical softcopy are performed over some x-ray CR images by using the display system. Using high-end monitors may decrease proposed method. Experimental results show that this those drawbacks. However, for those developing countries method can effectively reduce the quantum noise while enhancing the subtle details; the visual quality of X-ray CR such as China, the cost savings is needed in displaying images is highly improved. medical image. To this end, many hospitals are using the commercial displays for replacing medical grade displays. Index Terms—Wavelet denoising, image enhancement, These result in baddish quality in softcopy display. The unsharp masking visual quality of medical images is the key factor affecting diagnosis accuracy in PACS. Moreover, X-ray CR images are used widely in the clinical diagnosis.

This study is supported by the Project for Science and Technology Development of Zhengzhou (No.10PTGG399-2)

Therefore, it is necessary to improve the visual quality of Third, contrast is another important factor for X-ray CR images. diagnostic reading, medical grade display offer a better Image enhancement is a very powerful tool to improve contrast ratio than the commercial display, which enables the visual quality of images. Numerous enhancement them to be capable of rendering more DICOM JNDs. methods have been proposed in the literature. Mainly on Fourth, DICOM3.14 standard proposes a Grayscale grayscale transform and frequency domain transform. Standard Display Function (GSDF), which is much better Windowing/ leveling and the histogram equalization for viewing medical images. However, most commercial approach are two of the most widely used spatial uniform displays have their own luminance response functions, contrast-based enhancement technique, which are simple which do not follow DICOM standards. global enhancement approach and observers can Moreover, the consistency of display images over time interactively change the image contrast in any grey level and temperature is very significant in medical imaging; range [2]. Studies of frequency domain transform mainly medical displays can continuously measure the brightness concentrate on the wavelet transform [3]-[4]. However, of the display and even correct it if necessary. However, artifacts are inherent in these enhancement processes, such the commercial displays, its brightness is variable as the as, at bone-soft tissue boundaries. These artifacts have a change of time and temperature. detrimental effect on diagnoses. In some cases they can be On the other hand, noise remains in the original images pathological evidence in a normal radiograph, or they can also affect diagnosis accuracy for softcopy display. hide subtle lesions. Especially, an important issue in X-ray Our research purpose is to improve the diagnostic value CR images is to acquire high quality images with the of the images by image processing; enabling commercial lower clinical dosages to eliminate the potential danger of displays can provide long-term image viewing of the radiation. However, simply downscaling the radiation highest quality for efficient diagnosis and referral. intensities over the entire scan results in increased quantum noise. With the detection of the statistical nature III. PROPOSED ALGORITHM of photon, it is found that Poisson noise is one of the Our proposed method consists of the following steps. important factors degrading X-ray CR images at low Fig. 1 shows the flow chart of the image processing. count level. Several methods with a discrete wavelet transform have been proposed to remove the noise [5]-[6]. Such as Donoho’s soft thresholding algorithm [5], Zhang’s algorithm [6], etc. most of the previously developed wavelet-based denoising algorithms are to remove additive Gaussian noise. However, for removing Poisson noise, these de-noising techniques needs further to be researched. In this paper, a wavelet-based medical image denoising and enhancement algorithm was presented. We first investigated a wavelet-based de-noising algorithm. Then we modify the image enhancement technique in [7] and propose a new scheme for medical image de-noising and enhancement. By using the proposed method, the quantum noise on x-ray CR images can be reduced and the images can be enhanced effectively. Consequently, the visual quality of X-ray CR images is highly improved.

II. FACTORS FOR AFFECTING THE VISUAL QUALITY Figure 1. Flow chart of the proposed method Typically there are two factors affecting the visual Step1: Design a wavelet-based filter for removing quality of x-ray CR images displayed in PACS. They are quantum noise remains in the original images. the used display and the original image quality. Step2: Enhance the image contrast using Gamma Compared to medical grade displays, the commercial Correction. displays have some disadvantages for showing medical Step3: Extract the high frequency information of the imaging. original image. First, the commercial displays have a smaller Step4: Classify the high frequency information. luminance range, which enables them to offer the smaller Step5: Use image fusion to obtain a processed image. luminance difference versus between Just-Noticeable Step6: Calibrate the used display on the basis of DICOM Difference (JND) indexes. Such smaller luminance GSDF. difference is hard to be discerned by human eye. Step7: Decide parameters of the image fusion. Second, medical image is 10 bit or 12 bit. The number of available shades of gray on most commercial displays is limited to 256 (8 bit) only, enabling them hard to render every grayscale as defined by medical image.

F andG represent high-pass and low-pass filter at scale j, A. A wavelet-based filter j j Noise remains can mask and blur important features in resulting from interleaved zero padding of filters and (j>1). denotes the original image the real medical images. Fig.2 is an example of shown Fj−1 G j−1 cA0 images and the output of scale j,cAj would be the input of scale j+1. denotes the low-frequency (LF) estimation after cAj+1 the

Figure 2. A display image with big nodule using the commercial display. This image is degraded by noises as shown in Fig.2, enabling the detail of the texture Figure 3. Stationary wavelet decomposition of a two-dimensional image hard to be found. Thus, we first consider the method of de-noising enhancement. stationary wavelet decomposition, while h , v and cD j+1 cD j+1 In the reduction of noise, we have to remove the noise cDd denote the high frequency (HF) detailed information component appropriately while retaining the edge j+1 components; especially, the ribs and the blood vessels can along the horizontal, vertical and diagonal directions, be viewed as potential edges, which are important features respectively. These sub-band images would have the same in medical images. Because the spectrum of their features size as that of the original image because no down- is very similar to that of the noise, this is the difficulty of sampling is performed during the wavelet transform [15]. image processing. This kind of two-dimensional SWT leads to a Because wavelet transform are capable of trading one decomposition of approximation coefficients at level j in type of resolution for the other, which makes them four components: the approximation at level j+1, and the especially suitable for the analysis of nonstationary details in three orientations (horizontal, vertical, and signals. One privileged area of applications where these diagonal). Conversely, through the inverse wavelet properties have been found to be noise reduction. Our transform, the reconstructed image was obtained. previous research has proved that the wavelet-based de- Based on the wavelet thresholding scheme and SWT, noising techniques perform better than the optimal space- an improved wavelet filter is proposed. The proposed invariant solution (Wiener filter) [8]- [9]. Therefore, we method consists of the following steps: choose the wavelet-based de-noising method in this paper. Step1: Choose the best suitable wavelet and level based One of the widely discussed techniques is the wavelet on theory analysis. thresholding scheme, which recognizes that by performing Since Haar wavelet is self-reproducing across scales, it a wavelet transform of a noisy image, noise will be is an appropriate transform to the case of the Poisson represented principally as small coefficients in the high distribution. Moreover, in order to satisfy real-time frequencies. Thus in theory a thresholding, by setting application, the lower level is chosen to decompose an these small coefficients to zero, will eliminate much of the image. Here, two levels are chosen. noise in the image [10]. Step2: Decompose the noisy image to some subband Using Discrete Wavelet Transform (DWT), many images using SWT. methods for remove additive Gaussian noise have been SWT overcomes the drawback of the classical DWT, proposed [11]-[14]. However, based on the classical DWT which is a time-invariant transform. Therefore, we select de-noising provides multi-scale treatment of noise, down- SWT in this study. sampling of sub-band images during decomposition and Step3: The universal threshold is calculated by the the thresholding process of wavelet coefficients may following Donoho’s method for denoising automatically. cause edge distortion and Gibbs ring artifacts in the It can be exp-ressed as: reconstructed images [13]-[14]. In order to overcome this drawback, we choose Stationary Wavelet Transform T = σ 2 N)ln( , (SWT). whereσ = Median abs(wi 6745.0/))( is the SWT is proposed on the basis of orthogonal wavelet transform, and it remove the down-sampling in the estimated noise variance, wi is the detail coefficients of decomposition as illustrated in Fig.3. In Fig.3, SWT, and N is the sampling length of noisy image. cA0 represents a two-dimensional original image.

Step4: The Donoho’s soft thresholding method is applied to the details to modify SWT coefficients. Step5: The denoised image is reconstructed by the inverse SWT. In order to evaluate the de-noising method, we perform a series of de-noising experiments using images with controlled (a) (b) (c) noise amounts. Results of quantitative evaluations are Figure 5. (a) Noisy image (PSNR =24.37), (b) The proposed method presented. (PSNR =38.35), (c) DWT method (PSNR = 26.19). Images with Poisson noise are created using the following TABLE 1. PSNR MEASUREMENTS OF IMAGES method. Image Amount of Noise DWT The proposed No. noise Image method method The noisy image g(x ,y ) is generated by multiplying 1 λ = 02.0 15.54 25.14 29.95 the original pixel values by λ and using the obtained λ = 07.0 21.05 33.81 36.55 results as input to a random number generator which 23.41 36.75 38.47 returns truncated to the 12-bit gray-scale digitization λ = 2.0 range. 2 λ = 02.0 15.02 27.56 28.74 λ = 07.0 19.24 34.03 35.82 1 λ = 2.0 23.97 36.46 37.07 yxg ),( = λ yxfPoission )),(( (1) λ 3 λ = 02.0 14.33 28.60 30.05 λ = 07.0 20.82 33.74 35.85 f(x ,y )and gxy(, )are the gray-scale values in the λ = 2.0 25.70 35.95 37.67 original and noisy images, respectively. Fig.4 shows the 4 16.25 29.49 31.75 simulation of this process. λ = 02.0 λ = 07.0 21.13 33.95 36.83 λ = 2.0 24.81 37.28 39.29 Table 1 and Fig.6 show PSNR comparisons of the proposed method with the DWT method. The yellow line shows the PSNR of the denoised images by the proposed method. It can be noticed that the images that were Figure4. Poisson noise simulator applied the proposed method showed the best results.

With this approach, we maintain a constant mean PSNR Comparisons grayscale value at different noise levels. The amount of noise depends upon λ . 60 We use Peak Signal-to-noise (PSNR) as the evaluation Original image criterion of denoised images. The PSNR is calculated 40 using Equation (2) PSNR 20 DWT method S 2 PSNR=10 log (2) 0 10 MSE The proposed 1357911 method 1 N M Where, MSE= ∑∑ − mnfmng )],(),([ 2 (3) Image No. MN n 1 m== 1 S is the maximum pixel value, and PSNR is measured in decibel (dB). Figure 6. PSNR Comparisons We execute the algorithm to the obtained noisy images. Also, we evaluate the performance of the proposed For these noisy images, the PSNR was estimated before algorithm on real CR images with inherent noise. and after application of the proposed algorithm. The original and denoised images are shown in Fig.7 Experiment results are shown in Fig.5. Fig.5 (a) shows a (a), (b), respectively. Let us look round Fig.7 (a) and noisy version (PSNR =24.37), Fig.5 (b) shows the image Fig.7 (b). As one can see, the denoised image maintained (PSNR ==38.35) denoised by our technique, and Fig.5(c) sharpness and many fine details, while the noise was shows the image (PSNR =26.19) denoised by the DWT removed. It is visually better than the original image. method.

extracted high frequency components or the size of graininess is smaller, it is considered as noise. Secondly, we set a threshold T by analyzing this absolute value. Thirdly, we construct three temporary images according

to the following rule. If fh (,xy )> T, fh (,xy ) is

classified as the second temporary image f2 (,xy ), this is the diagnosis feature component in the high signal range.

Else if fh (,xy )< − T, fh (,xy ) is classified as the third temporary image f (,xy), this is the diagnosis feature 3

(a) The original image (b) The denoised image component in the low signal range, else fh (,xy ) is Figure 7. Comparison of the original and denoised image classified as the fourth temporary image f4 (,xy ), this is B. Contrast enhancement the noise component. Since x-ray CR images inherently have a wide E. Image Fusion dynamic range up to 10 bits or 12 bits and low detail On the basis of above four obtained temporary images, contrast. Moreover, diagnosis feature such as nodules and the enhanced image is reconstructed using image fusion as edges are often showing up as relatively low-contrast shown in Equation (6). white circular objects. In order to enhance low contrast white circular objects, we first compress low signal fe(,xy )= fxy1233 (, )+∗ afxybfxycfxy (, ) +∗ (, ) +∗ (, )(6) components in the denoised image fd (,xy )using Where, f1(,xy )is obtained using (2), γ is the gain Gamma correction, so that diagnosis feature can be factor for adjusting the compression ratio of non- enhanced naturally. diagnosis feature component. a is the gain factor for γ f1(,xy )= fd (, xy ) (4) adjusting the enhanced diagnosis feature component in the high signal range. b is the gain factor for adjusting Where, f1(,xy ) is the first temporary image. It is the enhanced diagnosis feature component in the low obtained using Equation (4). The γ value can be adjusted signal range. c is the gain factor for adjusting the by user. removed noise component. C. Extracting F. Calibration The High Frequency Component can be extracted In displaying a medical image, image data will be according to the following steps. transformed according to the corresponding Lookup Table step1: Design a lowpass filter using Gaussion function, (LUT) on monitor, which is the final link between the where the designed lowpass filter is an integer vector with image data and the eye brain system of human observer size 33. [16]-[19]. DICOM 3.14 standard proposes a Grayscale step2: The designed lowpass filter is used as Standard Display Function (GSDF), which is much better convolution kernel to compute a very blurred version of for viewing medical images. However, the LUT in the general display is preset for normal text and picture the original image yxf ).,( o display but not for medical softcopy image display, the

The obtained blurred image is defined as b yxf ).,( general display’s luminance response curve does not Step3: The high frequency component is computed follow GSDF defined as DICOM3.14. Thus, to make the using medical softcopy image display system offering full Equation (5). compliance with DICOM GSDF [18]-[19], we need to calibrate LUT in the general display. Moreover, the fhob(,xy )=− f (, xy ) f (, xy ) (5) monitor’s luminance response curve will appear degradation after a period of time; therefore, the Where, h yxf ),( is the extracted high frequency calibration process need be performed routinely. component. G. Deciding Parameter D. Classification In this study, the luminance response function used We can implement the enhancement of high-frequency commercial display is calibrated firstly according to the details of medical images in either spatial domain or GSDF defined in DICOM3.14 standard. Then frequency domain. Because both diagnosis feature and parameters γ , a, b, and c are decided by user while noise are high frequency components, the drawbacks of existing methods are enhancement of noise present in viewing the image quality. Also, we will make DICOM images. In order to solve this problem, we analyze and 3.14 calibration and update parameters to keep the image classify the extracted high frequency components into consistent over time. three feature images. Firstly, if the absolute value

Therefore, although the commercial display is used, the IV. EXPERIMENT AND EVALUATION enhanced image can provide long-term image viewing of the highest quality for efficient diagnosis and referral. We first perform a series of experiments using images in the Standard Digital Image Database, Japan: nodule V. CONCLUSION 154, and non-nodule 93 on chest radiogram, to verify this technique. A novel method to improve the visual quality of x-ray Fig.8 (a), (b), (c), and (d) show the original images with CR images has been presented. We have demonstrated big, middle, small nodules, and non-nodule using the that the proposed technique not only can offer effective commercial display respectively. The enhanced results of noise removal in noisy medical images and enhancing these images are shown in Fig.9 (a), (b), (c), and (d) sharpness, but also can improve the diagnostic value of respectively. The images displayed in Fig.8 present the shown image on the commercial display successfully. undesired properties like low signal-to-noise (SNR) and low contrast ratios. The displayed images have so poor image quality that nodules can not be seen in these images. The enhanced results of these images are shown in Fig.9. Big nodule and middle nodule not only can be seen clearly in Fig.9 (a) and (b); but also, little nodule in Fig.9 (c) and the detail feature in Fig.9 (d) can be observed. The results show

Figure 9. Show the enhanced images using the commercial display

REFERENCES [1]Liu Jiquan; Feng Jingyi; Lao Duchun; “DICOM GSDF Figure 8. Show the original images using the commercial display Based Cal- ibration Method of General LCD Monitor for Medical Softcopy Ima- ge Display”,Bioinformatics and st that the diagnosis feature has been enhanced effectively in Biomedical Engineering, 2007, ICB- BE 2007, The 1 International conference on Digital object identifier: the processed images. 10.1109/ICBBE.2007.298, Publication Year:2007, pp.1153- Further more, we have evaluated this technique based 1156. on minimum standards to meet the American Association [2] Iyad Jafar, Hao Ying, “A New Algorithm for Adaptive of Physicists in Medicine (AAPM) and Japan Industries Contrast Enhancement Based on Human Visual Properties for Association of Radiological Systems Standards (JESRA). Medical Imag- ing Applications,” IJCSNS International Journal In visual check, acceptance test result report has of Computer Scie- nce and Network Security, VOL.7 No.7, July indicated that the patches’ luminance differences among 2007. 16 steps can be clearly recognized, 5% and 95% patches [3] Jiao Feng, Naixue Xiong and Bi Shuoben. “X-ray Image can be visible, the judgment-use positions on the reference Enha-ncement Based on Wavelet Transform”, Proceedings of IEEE Inte-rnational Conference on Asia-pacific service clinical image can be visible without any problem, smooth, computing, pp. 1568-1573, 2008. stable and continuous display can be presented, and [4] Miller, M.; Kingsbury, N. “Image denoising using derotated artifacts do not be present. co-mplex wavelet coefficients”, IEEE Transactions on Image The measured values, such as contrast response, Processing, Vol. 17, pp. 1500-1511, 2008. luminance deviation, luminance ratio, etc, are the range of [5]D.L. Donoho, “De-noising by soft-thresholding,” IEEE acceptance in AAPM and JESRA Standards. Trans. Inform. Theory, vol. 41, No.3,pp.613-627, 1995.

[6] Zhong J, Ning R, and Conover D. Image denoising based on Zhengzhou University. Her current research interests are multiscale singularity detection for cone beam CT breast wavelet applications in signal processing, medical image imaging. IEEE Trans MedImaging, 23: 696-703, 2004. diagnosis system, three-dimensional image reconstruction and [7] H.Jiang, Y. Lu and L. Ma. “A Novel Image Enhancement image processing. She is a member of Research Institute of Tech-nique for CR Chest Images,”2010 RISP International Signal Processing, Japan. Workshop on Nonlinear Circuits and Signal Processing, Hawaii, USA, March 3-5, 2010, pp..239-242. Zhongyong Wang received his B. S. [8] H. Jiang, T. Yahagi, T. Okamoto, and J. Lu,“An Adaptive and M. S., degrees in Automatic Control from Deno- ising Method for Medical CR Images Using Wavelet Harbin Shipbuilding Engineering Institute, Packet”, Journal of Signal Processing, vol.7, November Harbin, China, in 1986, 1988, respectively, 2003,no.6, pp. 443-452. and received his Ph. D. degree in Automatic [9] J. Lu, H. Jiang, H. Jiang, and T.Yahagi ,``Restoration of Control Theory and Application from Xi'an Images with Poisson Noise Based on Wavelet Shrinkage Jiaotong University, Xi'an, China, in 1998. Technique'', 2004 RISP International Workshop on Nonlinear Since 1988, Zhongyong Wang has been with Zhengzhou Circuits and Signal Proc- essing(NCSP'04), Hawaii, USA, University, Zhengzhou, China, as a Lecturer in the Department March 5-7, 2004,pp.391-394. of Electronics. From 1999 to 2002, he was an Associate [10] S. Mallat, A Wavelet Tour of Signal Processing. Academic Professor, and in 2002 he was promoted to Professor in the Press, London, 1998. School of Information Engineering. He is currently the Dean of [11] G. Chang, B. Yu, M. Vetterli, “Adapting to unknown the School of Information Engineering and a member of Digital smooth- ness via wavelet shrinkage,” IEEE Trans. Image. Proc., Imaging Technology Research Center , Zhengzhou University. vol. 9, pp. 1532 -1546, 2000. Prof. Wang’s general fields of interest cover numerous aspects [12] Zhong J, Ning R, and Conover D. “Image denoising within embedded systems, signal processing and based on multiscale singularity detection for cone beam CT communication theory. breast imaging,” IEEE Trans. Med. Imaging, 23: pp.696-703, 2004. Ling Ma received his M.S. degree in [13]Chun Jiao, Dongming Wang, Hongbing Lu, Zhu Zhang, computed mathematics from Hebei Normal Jerome Z. Liang,“Multiscale Noise Reduction on Low-Dose CT University, P.R. China and Ph. D. degrees in Sinogram by Stationary Wavelet Transform”, 2008 , IEEE Department of Manufacturing Engineering Nuclear Science Symposium Conference Record, pp.5341-5344. from Beijing university of Aeronautics [14] Zachary S. Kelm, Daniel Blezek Ph.D, Brian Bartholmai &Astronautics,. P.R. China, in 1987 and M.D. “Optimizing Non-local Means for Denoising Low Dose 1997, respectively. In 1998, he joined Tokyo CT”, July 1 2009, pp.662-665. Research Institute of Tani Electronic Company, Tokyo, Japan, [15] Nason GP, Silverman BW. The stationary wavelet as a researcher. Presently He is working as a Research Fellow transform and some statistical applications in wavelet and in Fast Corporation in Japan. His research interests include statistics. In:Antoniadis A ed.Lecture Notes in Statistics. Berlin: image processing, pattern recognition, 3D machine vision and Spinger Verlag, 281-299, 1995. computer graphics. [16] Carrino J., “Image Quality: a clinical perspective”, In: Siegel E, Reiner B. and Carrino J., SCAR University Primer 3: Quality Ass- urance in the Digital Medical Enterprise, Society Yumin Liu received her M. S. degree for Computer A- pplications in Radiology, 2002. in mathematics from Zhengzhou University [17] J.Wang and Q.Peng, “An Interactice Method of Assessing and Ph.D degree from Nankai University, in the Characteristics of Softcopy Display Using Observer 1988 and 2003 , respectively. In 1988, she Performance Tests”, J. of Digital Imaging, Vol. 15, Suppl. 1, pp. joined Zhengzhou Institute of Aeronautics, 216-218, 2002. China, and in 1995 she was promoted to [18] E. Samei, A. Badano, D. Chakraborty, et al, AAPM Online Professor in the Dept. of Application Science Report No.03, “Assessment of Display Performance for Zhengzhou Institute of Aeronautics .From 2001 to Current, she Medical Imaging Systems”, 2005. is a Professor in Zhengzhou University. She currently is a [19] David M. Jones, “Utilization of DICOM GSDF to Modify member of Digital Imaging Technology Research Center, Lookup Tables for Images Acquired on Film Digitizers”, J. of Zhengzhou University. She current research interests are in Di-gital Imaging, Vol. 19, No. 2, pp. 167-171, 2006. Quality Engineering, Statistical Technique, Corporate Governance Customer Satisfaction Measurement, Six Sigma Huiqin Jiang received her M.S. Management and image processing. degree in mathematics from Zhengzhou University, P.R. China and Ph.D degree in information Science from Chiba University,. Ping Li received the bachelor degree Japan, in 1988 and 2004, respectively. In 1988, in electronic and information engineering she joined Zhengzhou University. From a from the PLA Information Engineering Lecture to Associate Professor, she has University. In 2010, she joined Zhengzhou worked for ten years in Zhengzhou University. In 1999, she University, as a master student of the school gone to Japan for study and work. She worked as a Research of information engineering. Her current Fellow in TERARECON Inc., USA and RealVision Inc., Japan, research interests are image compression and respectively. In 2009, she was promoted to Professor in the image processing. She is a member of school of information engineering, Zhengzhou University, and Digital Imaging Technology Research Center, Zhengzhou created the digital medical image technique research center, University.